Start-kit & baselines.
The start-kit is a set of guides in the NeuroAI repository. This page points you to them and lists the baseline scores to beat.
The rules are not on this page. Submission caps, eligibility, external data, compute, the reproducibility audit and authorship all live on the Rules & FAQ page. Read them before your first upload.
Train, evaluate locally, submit.
The guides live in the NeuroAI repository and are published in the NeuralBench documentation. They show how to 1) train a baseline, 2) evaluate it locally with the official evaluator, and 3) submit a model. Start with the overview, then open the guide for your track.
neuralbench eeg image
EEG-to-Image
THINGS-EEG and Alljoined; top-5 retrieval accuracy in a frozen DINOv2 embedding space.
Open guide →
Track 2 · neuralbench eeg motor_imagery
EEG-to-BCI
Stieger2021 as the stand-in cohort; balanced accuracy, within-user and cross-session.
Open guide →
Track 3 · neuralbench eeg sleep_onset
Sleep onset
Sleep-EDF as the stand-in until the Muse cohort drops; binned MAE on onset latency, in seconds.
Open guide →
Track 4 · neuralbench emg pose -m neuropose
EMG-to-Pose
Salter2024 EMG2Pose: 16-channel wrist sEMG to 20 joint-angle trajectories, evaluated in degrees.
Open guide →
Pre-computed baselines on representative public datasets.
Numbers below are NeuralBench replications on representative public datasets, mean ± std, taken from the competition proposal's Table 1. They are the warm-up targets to beat. The official warm-up split scores arrive Sep 16, 2026.